Skip to content
SIH Buddyby Ganeev Singh
Dev
All problem statements
SIH26053Strong pickacceptance 4/5

Adaptive Variable Resolution 2.5D Lidar Mapping for Dynamic Environment Perception

DRDO · Transportation & Logistics · Software

Public data, numeric targets, a measurable memory claim and a genuinely novel representation make this among the cleanest statements on the portal — spend your time on the variable resolution grid rather than the segmentation network, because that is where both the difficulty and the credit are.

What it actually is

A LiDAR throws out millions of points a second, which is too much to process live, but flattening it into a plain 2D map throws away the height information you need to see a kerb or a low branch. The ask is a middle path modelled on how eyes work — full detail close to the vehicle where safety depends on it, and progressively coarser further out. The map should also know what it is looking at, not just how tall it is.

What to build

A perception pipeline turning raw point clouds into a semantically labelled elevation map at non-uniform resolution: a segmentation network classifying every point into drivable terrain, static obstacles such as walls and poles, and dynamic objects such as pedestrians and vehicles; a variable resolution grid engine projecting those classified points into a 2.5D structure at the resolutions the statement specifies — roughly five centimetre cells within ten metres coarsening to fifty centimetre cells out to a hundred — handling the alignment and data loss problems that arise when cells of different sizes meet; a real-time visualisation colour-coding terrain and object classes; and a benchmark reporting frame rate, classification accuracy broken down by distance band, and memory footprint against a uniform high-resolution baseline.

Smallest thing that wins the room

Play a driving sequence live with the foveated grid visibly coarsening outward, a pedestrian tracked in the fine zone, and a memory counter showing your map using a fraction of what the uniform 3D representation costs at the same frame rate.

How crowded this one gets

A guess, projected from the 2025 statements — the last year where both the submission counts and the winners were published.

Busy240–500 teams expectedroughly 1 in 201–421 wins it

Quieter than 16% of the 226 · #191 of 226 by expected field · reaches the 500 cap

Busier than most. Expect several teams to arrive at the same obvious solution.

Why: defence, intelligence and space bodies drew small fields.

This is a guess, not a fact

Nobody has published 2026’s numbers yet. This is an analysed estimate from last year’s pattern, so please do not take it as the truth — check the live counter on the SIH portal before you decide anything. The range covers the middle half of likely outcomes, so one statement in two lands outside it. Entry closes at 500 ideas per statement, so no range goes past that — a statement that reaches the cap fills and shuts rather than drawing an unlimited crowd. The model reads only three things a team can see before choosing — software or hardware, the theme, and what kind of body posted it — and those explain about a quarter of the variation in last year’s field sizes (R² 0.25 on held-out statements). Trust the band more than the number, and the ordering more than either. It cannot see how good your idea is, which is the part that actually decides it.

The scores

The number is the shorthand. The line under it is the reason.

What you will be writing

  • MinkowskiEngine or SPVNAS sparse convolutional segmentation
  • PointNet++ point cloud feature extraction
  • quadtree-backed variable resolution elevation grid
  • SemanticKITTI / nuScenes labelled point clouds
  • Open3D or rerun.io real-time visualisation
  • memory footprint benchmarking against uniform voxel grid
  • LiDAR perception
  • Autonomous navigation
  • Spatial data structures

Prior art to read before you start

point cloud semantic segmentation · variable resolution occupancy mapping · real-time perception memory optimisation

Analysed by Claude Opus. Every score above is a judgment call with its reasoning attached — kindly cross-check this against the official statement on the SIH portal before your team commits to it.